This repository shares Maimemo’s report “A Study of Memory Algorithms 1.0” (2021-07-06), which documents the objectives, methodology, and outcomes of the team’s memory-algorithm research.
The report is provided in both Chinese and English. Many thanks to the AI translation tools that helped prepare the bilingual editions for reference.
A Study of Memory Algorithms 1.0 (Chinese version).pdf: The original Chinese report.A Study of Memory Algorithms 1.0 (English version).pdf: The English translation.README.zh.md: A Chinese README that mirrors this summary.
According to Section 1.3 of the report, the work contributes:
- A comprehensive review of major memory algorithms (Ebbinghaus, Maimemo, SuperMemo/Anki, Duolingo) and their evolution.
- A theoretical investigation of memory stability, retrievability, review scheduling, and evaluation metrics.
- Empirical validation using Maimemo’s Full Study Record dataset, reproducing the forgetting curve and stability growth phenomena.
- Design updates to Maimemo’s algorithmic matrices (FVI, MMDB) and development of MMDB 2.0.
- Enhancements to long-term and same-day review algorithms, leading to Maimemo Memory Scheduling Algorithm 2.0.
- Supplementary engineering work, including database and synchronization improvements plus new statistical visualizations.
Chapter 3 notes several inconsistencies between the observed data and SuperMemo’s published results:
- The forgetting curve fitted to FSR data introduces a parameter
a, producing a non-zero long-term recall limit, unlike Dr. Piotr Wozniak’s theoretical curve that decays to zero. - Stability-increase experiments yield trends that differ from SuperMemo’s reports: the effect varies with retrievability and peaks near 80%, with further data collection required for confirmation.